Gemini 3.7 Flash launch brings coding gains and a temporary API rate

Google’s Gemini 3.7 Flash launch puts a new model for coding, agentic workflows and enterprise tasks into its developer and enterprise products, three weeks after Gemini 3.6 Flash. The release matters less as a new model name than as a short-cycle upgrade with an introductory API rate that is scheduled to double on Jan. 1, 2027.

Google announced the model on Aug. 13 and said developer feedback and algorithmic improvements informed the release. The company positions 3.7 Flash for software engineering, web development, knowledge work and enterprise workflows, and says it has improved multi-step planning and tool calls. Those are Google’s product claims, and the available evidence does not include independent performance testing.

What does Gemini 3.7 Flash cost?

Google lists introductory prices through Dec. 31, 2026 of $0.75 per million input tokens and $3.75 per million output tokens. From Jan. 1, 2027, the listed rates rise to $1.50 for input and $7.50 for output per million tokens.

For a workload consuming 1 million input tokens and 1 million output tokens, that works out to a listed $4.50 through year-end and $9.00 after the scheduled increase, before applicable discounts or other charges. Teams evaluating the model for high-volume agents should model the post-promotion rate rather than treat the current pricing as permanent.

Google reports better coding and automation results than 3.6 Flash

Google’s published evaluation table reports gains against Gemini 3.6 Flash across several tests. On FrontierCode 1.1 Main, a production-code-quality benchmark, 3.7 Flash scored 43.6%, compared with 34.4% for its predecessor. On DeepSWE v1.1, Google reported 65.3%, versus 48.6% for 3.6 Flash, while GPT-5.6 Terra scored 69.6% in the same table.

Google also reported a 1,588 Elo score on Code Arena’s web-development evaluation, up from 1,538. For enterprise-oriented tasks, its table lists 30.4% on AutomationBench, compared with 17.0%, and 34.0% on GDP.pdf document comprehension, compared with 22.0%.

The comparisons do not make 3.7 Flash a universal leader. Google’s own table lists competitors ahead on measures including DeepSWE, Terminal-bench tests, OSWorld-2.0 and Agent’s Last Exam. The practical claim is narrower: Google reports stronger results than 3.6 Flash on several coding and automation benchmarks, with performance varying by workload.

Where can developers use Gemini 3.7 Flash?

According to Google’s model card, the model is distributed through the Gemini API, Google AI Studio, Google Antigravity, Gemini App-Spark, Gemini Enterprise App and Gemini Enterprise Agent Platform.

It accepts text, images, audio and video, supports a context window of up to 1 million tokens and produces text outputs of up to 64,000 tokens. The model is based on Gemini 3.6 Flash and includes configurable thinking settings intended to trade among output quality, cost and latency. For buyers, that makes the relevant test a production workload: benchmark uplift is useful, but token economics and agent reliability will determine whether the upgrade changes total operating cost.

This story draws on original reporting from SiliconANGLE.